Computational Aesthetics of Image Classification and Evaluation

Assessing image aesthetic appeal referring to principles of the nature,visual appreciation and psychology of human beings,is a highly subjective task.Computational image aesthetics aims to assess the aesthetics of images automatically.In this paper,we design a comprehensive method for aesthetic evaluation focusing on digital images.First,key regions of images are extracted through salient region detection method.Then a set of discriminative low-level visual features and high-level aesthetic features from the global images are extracted,as well as regional features that characterize the key regions.Based on the extracted features and human aesthetics ratings,the image aesthetic classifier and the aesthetic score prediction model are built through machine learning methods.As a result,the aesthetic category(highly aesthetic VS low aesthetic),as well as the aesthetic score,of an image can be evaluated automatically through our models.The experiment results demonstrate that our image aesthetic classifier achieves a promising classification accuracy of 75.37%.In the mean time,our aesthetic score prediction model got a correlation coefficient of 0.790 and RMSE of 0.244 on images' aesthetic score between automated assessment and human beings' subjective aesthetics evaluation.The results show that the method proposed in the paper is competitive with previous approaches.